319-1040-2-PB.pdf

8/9/2019 319-1040-2-PB.pdf

1/14

Journal of Agriculture and Rural Development in the Tropics and Subtropics

Vol. 111 No. 2 (2010) 129-142

urn:nbn:de:hebis:34-2011052737598 ISSN: 1612-9830 journal online: www.jarts.info

Assessment of land use and land cover changesduring the last 50 years in oases and surrounding rangelands

of Xinjiang, NW China

Andreas Dittrich , Andreas Buerkert , Katja Brinkmann*

Organic Plant Production and Agroecosystems Research in the Tropics and Subtropics,

University of Kassel, Witzenhausen, Germany

Abstract

An analysis of historical Corona images, Landsat images, recent radar and Google Earth images was conducted

to determine land use and land cover changes of oases settlements and surrounding rangelands at the fringe of the

Altay Mountains from 1964 to 2008. For the Landsat datasets supervised classification methods were used to test

the suitability of the Maximum Likelihood Classifier with subsequent smoothing and the Sequential Maximum A

Posteriori Classifier (SMAPC). The results show a trend typical for the steppe and desert regions of northern China.

From 1964 to 2008 farmland strongly increased (+ 61 %), while the area of grassland and forest in the floodplains

decreased (- 43 %). The urban areas increased threefold and 400 ha of former agricultural land were abandoned.

Farmland apparently affected by soil salinity decreased in size from 1990 (1180 ha) to 2008 (630 ha). The vegetated

areas of the surrounding rangelands decreased, mainly as a result of overgrazing and drought events.

The SMAPC with subsequent post processing revealed the highest classification accuracy. However, the specific

landscape characteristics of mountain oasis systems required labour intensive post processing. Further research is

needed to test the use of ancillary information for an automated classification of the examined landscape features.

Keywords: Altay Mountains, LUCC, Landsat, Corona, NDVI, SMAPC, MLC

Abbreviations:

SMAPC = Sequential Maximum A Posteriori Classifier

NDVI =Normalized Difference Vegetation Index

MLC =Maximum Likelihood Classifier

LUCC= Land use and land cover changes

GCPs = Ground Control Points

1 Introduction

During the last decades many landscapes have un-

dergone large structural changes given the introduction

of new management practices and the social, political,

and economic framework controlling land use (di Cas-

tri & Hadley, 1988; Turner et al., 2001). In China,

such processes have reportedly resulted in widespread

land degradation, often caused by population growth

and over-intensification of agriculture such as excessive

Corresponding author

Fax+49 5542-98 1230; Email: [email protected]

application of mineral fertilizers, overgrazing of steppes

and excessive cutting of trees and shrubs for fuel wood

(Zhu & Chen, 1994), man-made salinisation and losses

of wetlands (Zhang et al., 2007). Climate change effects

(Hu et al., 2003) are also claimed as recent drivers of

land use change (Goaet al., 2002; Kleinet al., 2004).

In the arid parts of NW China the state and develop-ment of oasis systems are of particular importance given

their dominant role in many landscapes of the Xinjiang-

Uyghur Autonomous Region (Liuet al., 2010). During

the last 50 years the population of this part of China

increased from 4.7 to 18.5 Mio. people and the area

of cultivated land expanded from 1.5 to 3.4 million ha

(Wiemer, 2004). Collectivisation programs fostered the

conversion of grassland into cropland and of herders

into agro-pastoralists (Millward & Tursun, 2004; Chu-

luun & Ojima, 2002). As a result of poor reclamation

of land, decreasing pasture areas, inefficient livestock

husbandry systems and rising demand of meat due torapid population growth and changes in food consump-
http://nbn-resolving.de/urn:nbn:de:hebis:34-2011052737598http://nbn-resolving.de/urn:nbn:de:hebis:34-2011052737598

8/9/2019 319-1040-2-PB.pdf

2/14

130 A. Dittrich et al./J. Agr. Rural Develop. Trop. Subtrop. 111 - 2 (2010) 129-142

tion preferences, the grazing pressure on the remain-

ing rangeland areas increased dramatically (Liu, 1993;

Graetz, 1994; Li et al., 2008). Since the 1980s, most

of the oases in Xinjiang are characterized by mecha-

nized cropping systems that depend on an intensive use

of irrigation water accompanied by often insuffi

cientdrainage (Zhanget al., 2001; Bruelheide et al., 2003;

Graefe et al., 2004). As a consequence, land degra-

dation through salinisation and sand encroachment in-

creased (Brunner, 2005; Xiuling, 2001). In 2000 about

50 % of the total river water in Xinjiang was used for

oasis irrigation leading to a rise of salt content in rivers

and lakes (Liet al., 2001).

To detect and monitor land use and land cover

changes (LUCC) and ongoing desertification processes,

satellite remote sensing, in conjunction with geographic

information systems (GIS), has been widely applied and

been recognized as a powerful analytical tool (Xiaoa

et al., 2004; Jia et al., 2004; Zhanget al., 2007).

For China, however, most LUCC studies examined

steppe and desert regions and only rarely targeted com-

plex oasis systems (Hao & Ren, 2009; Luoet al., 2008;

Li et al., 2004). These studies were either based on

visual image interpretation or on automated classifica-

tion algorithms. Other remote sensing studies in China

have been conducted to monitor desertification pro-

cesses (Hill, 2001; Baoping & Tianzong, 2001), quan-

tify rangeland biomass production (Runnstrom, 2003),

map vegetation types (Wang et al., 2002), and to anal-

yse the temporal and spatial development of urban areas

(Junet al., 2006). In contrast to the described studies,our work focused on LUCC in complex oasis systems at

the foothills of the Altay Mountains. The objectives of

the study were (i) to determine LUCC of oasis systems

and their surrounding rangelands in the Altay Moun-

tains between 1964 and 2008 using remote sensing tech-

niques and (ii) to evaluate the effectiveness of auto-

mated procedures to determine land use and land cover

changes at a broader scale. In this context, the suitabil-

ity of the commonly used Maximum Likelihood Classi-

fier (MLC) with subsequent smoothing (Pal & Mather,

2003; Booth & Oldfield, 1989) and the rarely used Se-

quential Maximum A Posteriori Classifier (SMAPC,Bouman & Shapiro, 1994) were compared. The latter

approach is often more accurate when classifying dif-

ferent types of crops cultivated on homogenous fields

(McCauley & Engel, 1995).

2 Materials and Methods

2.1 Study area

The study area is located in Qinghe County in the

NW part of the Dzungarian basin in the Xinjiang Au-

tonomous Region of China (Figure 1). The basin is sur-

rounded by the Altay range in the north and the Tian-

shan mountains in the south. The 482 km2 study area

(901530 903030 E and 462430 464830

N) encompasses several oasis settlements at altitudes

between 1185 and 1890 m a.s.l, including the town of

Qinghe in the South and the village Buheba in the North

(Figure 2 and 3). From 1962 to 1997, the mean temper-ature in Qinghe county was 1.1C with a mean annual

precipitation of 40 80 mm (NCDC, 2009). Given the

short vegetation period, spring wheat (Triticum aestivum

L.) is the dominant crop, which, after sowing in April is

widely grazed to foster tillering. Intensive greenhouse

vegetable production as an alternative to the cultivation

of irrigated spring wheat only occurs in the immediate

proximity of or within oasis settlements. At present an

unknown number of (agro-) pastoralists practice sum-

mer transhumance, while the reminder reportedly man-

aging about 30-40 % of Qinghes livestock, remain in

the agricultural area throughout the year with daily graz-

ing in the neighbourhood of the oasis area.

Since 2002, the local government has implemented

a forest policy, reconverting degraded arable land into

pasture or forest. Under this policy so far, 8,600 ha have

been reforested, and 3,800 ha of the reclaimed poplar

(Populus sp.) forest area are protected from grazing by

fences. In addition, 40,000 ha of original forest in the

north of Qinghe county have been put under protection.

2.2 Data acquisition and pre-processing

To identify LUCC that occurred from 1964 to 2008

in the study area, different satellite image sources wereused, such as Corona, Landsat, Google Earth images,

and TerraSAR-X (Synthetic Aperture Radar) radar im-

ages (Table 1). The earliest data resulted from two

panchromatic Corona (KH-4B) images taken in 1964 by

the U.S Geological Surveys Earth Resources Observa-

tion and Science (EROS). During the Corona program

from 1960-1972, several camera systems, referred to by

their KEYHOLE (KH) designator, were used. Of these

the most advanced, KH-4B had its best resolution of

1.8 m at nadir. We further obtained cloud free Land-

sat 5 and 7 Level 1 terrain corrected images (L1T, res-

olution = 30 m) from EROS, which have radiometric,

geographic and topographic corrections. Since 2003,

Landsat 7 imagery has been interrupted and contains

data gaps due to the failure of the Scan Line Correc-

tor (SLC), which compensates for the forward motion

of the satellite (Chanderet al., 2009). Thus, an already

processed, gap-filled product was obtained to accurately

depict the land use and land cover conditions in August

2008.

All Landsat images covered the whole study area.

The two TerraSAR-X radar images displayed 85 % of

the oasis and 25 % of the surrounding area, while the

Google Earth Quickbird images covered a total of

89 % of the study area.

8/9/2019 319-1040-2-PB.pdf

3/14

A. Dittrich et al. /J. Agr. Rural Develop. Trop. Subtrop. 111 - 2 (2010) 129-142 131

Fig. 1: Location of the study area (482 km2) in the Xinjiang autonomous region of NW China.

Fig. 2: The oasis settlement Qinghe and its surrounding in June 1964 (Corona image) and August 2002 (Google Earth image).

8/9/2019 319-1040-2-PB.pdf

4/14


Fig. 3: The oasis settlement Buheba and its surrounding in June 1964 (Corona image) and

August 2002 (Google Earth image).

Table 1: Provider, date of acquisition and spatial resolution (m) of the satellite images used for LUCC analysis in the

Dzungarian river basin, NW China.

Satellite Provider Date of Acquisition Spatial Resolution (m)

Corona KH-4B U.S. Geological Surveys Earth

Resources Observation an Science22 June 1964 1.80-7.6

Landsat-5 U.S. Geological Surveys Earth

Resources Observation an Science7 September 1990 Band 1-5, 7: 28.5

16 August 1999 Band 1-5, 7: 30; Band 8: 15Landsat-7 U.S. Geological Surveys Earth

Resources Observation an Science 8 August 2008 Band 1-5, 7: 25; Band 8: 12.5

TerraSAR-X German Aerospace Center May 2007 0.75

20 August 2002

8/9/2019 319-1040-2-PB.pdf

5/14


All Landsat images have been co-registered to the

radar and the Google Earth images, which have been

correctly georeferenced according to ground control

measurements. The scanned Corona images were geo-

referenced using readily recognizable features such as

roads and stone formations as ground control pointsand verifying the results based on georeferenced topo-

graphic maps and satellite images (Number of GCPs: 33

and 25; Root mean square error: 0.00002 and 0.00014;

Transformation: Spline and 3rd Order Polynomial).

The available TerraSAR-X images were used to cre-

ate a high resolution image containing the spectral infor-

mation of the 2008 Landsat dataset to identify training

and validation sites for the subsequent classification pro-

cesses. One common problem in images acquired with

the SAR technique, is the so called Speckle Effect hin-

dering a proper interpretation of these images (Oliver &

Quegan, 2004). For speckle noise reduction a combina-

tion of the Enhanced Frost Filter and the Local Sigma

Filter was applied using ENVI 4.2. (Research Systems,

Boulder, CO, USA). The fusion step of the HSV trans-

formed Landsat colour composites and the TerraSAR-X

images were processed based on the Principal Compo-

nent technique (Pohl, 1999) using ERDAS IMAGINE

8.3. (Leica Geosystems GIS & Mapping LLC., Nor-

cross, GA, USA).

2.3 Data analysis and classification methods

For the supervised classification of the Landsat

datasets a Maximum Likelihood Classifier (MLC)with subsequent smoothing and a Sequential Maximum

A Posteriori Classifier (SMAPC) were compared using

the open source software Quantum GIS 0.11.0 Metis

with the GRASS plugin (Quantum GIS Development

Team, 2009). The MLC is a simple pixel based method

which is often used for land use change detections (Pal

& Mather, 2003; Booth & Oldfield, 1989). The subse-

quent smoothing reduces the so called salt and pepper

effect characterized by scattered single pixels belong-

ing to another class than the majority of the neighbour-

ing pixels (Lillesand & Kiefer, 2000) and was applied

to improve the classification results. In its classifica-tion the SMAPC also accounts for the spatial informa-

tion between neighbouring pixels and produces more

homogenous regions than obtained with the pixel-by-

pixel method which computes local image pyramids at

different scales (Bouman & Shapiro, 1994; McCauley &

Engel, 1995). At each scale the best segmentation is cal-

culated based on the previous coarser segmentation and

the observed data using a spectral class model known

as Gaussian mixture distribution (Bouman & Shapiro,

1994).

Based on field observations and visual inspections of

recent Google Earth images, different land use classes

were detected within the study area (Table 2). The class

Urban Areas included all man-made facilities such as

houses in the countryside and bigger settlements as well

as industrial facilities. Water bodies were defined as

class 2 and riverbanks as class 3. The latter were situ-

ated next to the water bodies, and were defined as a sep-

arate class to avoid misclassifications with other ground

objects such as pathways or eroded land. To observethe development of the agriculturally used land classes

4 to 6 were defined based on spectral and spatial in-

formation such as the appearance of irrigation channels

and the more or less rectangular shape of the fields as

additional criteria. These areas were further discrimi-

nated according to their status (harvested-not harvested-

degraded). Areas with soil degradation due to soil salin-

ity were identified to (i) investigate how many areas

were already degraded and (ii) determine the extent to

which degraded arable land was reclaimed into pasture

and forest land. The non-agricultural areas within the

floodplain of the oases are typically characterised by

woodlands (natural riverside forest ofPopulus diversi-foliaand plantations with Populus alba) and grasslands

used for haymaking and grazing.

The non-vegetated and/or unutilized areas within and

outside the oases were allocated to class 9 (Mountains)

and the vegetated areas in the proximity of the oases to

class 10 (Rangelands). The definition of these classes

allowed the analysis of grazing effects on rangeland

areas.

To determine specific spectral reflectance patterns for

the supervised classification (Lillesand & Kiefer, 2000;

Sabins, 1997), training sites were defined for each class

based on the pan-sharpened Landsat image of 2008.

Subsequently, a signature file with the specific spec-

tral attributes for each class was calculated. A com-

parison of the Landsat histograms of the three differ-

ent years (1990, 1999 and 2008), which display the dis-

tribution of the digital numbers (DN) representing the

spectral reflection/absorption of the landscape (Wilkie

& Finn, 1996), revealed spectral differences for the year

1990, whereas 1999 and 2008 were more or less similar

(Figure 4).

It was therefore necessary to redefine the signature

files for the older Landsat images by using additional

training sites for 1990 and 1999. These have been visu-ally identified using spectral information of pseudo-true

colour and standard false colour composites, whereby

the visual interpretation was calibrated based on al-

ready known spectral and shape characteristics (colour,

structure, size) of the year 2008. Finally, the mean

of the spectral reflectance patterns was calculated for

each class in 2008, 1999 and 1990 and compared among

years.

For the oasis area (class 1 to 9) a supervised classifi-

cation was conducted, whereby the surrounding range-

land area was further classified using the Normalized

Difference Vegetation Index (NDVI) based on the Land-sat images. The NDVI is the ratio of the reflectance in

8/9/2019 319-1040-2-PB.pdf

6/14


Table 2: Predefined land use classes, number and total area (ha) of training sites (TS) and validation plots (VP) for the supervised

classification of LUCC in the Dzungarian river basin, NW China.

Class Number Class Name Ground Objects Number and

size of TS

Number and

size of VP

1 Urban Areas single houses, settlements, cities,industrial facilities

428 (7.0) 393 (7.2)

2 Water Bodies main river, branches, reservoirs 90 (7.0) 50 (4.3)

3 Riverbanks sandy areas next to class 2 54 (1.9) 45 (2.9)

4 Agricultural LandHarvested arable land (signs of soil

cultivation) harvested 78 (32.9) 65 (21.7)

5 Agricultural LandCrops arable land (signs of soil

cultivation) not harvested 110 (29.3) 96 (27)

6 Agricultural LandDegraded

arable land (signs of soil

cultivation) degraded due to soil

salinity

68 (27.7) 46 (5.5)

7 FloodplainWood woodland in the floodplain 69 (24.3) 66 (26.0)

8 FloodplainMeadow grassland and bushes in the

floodplain 49 (39.7) 51 (15.8)

9 Mountains non-vegetated areas, blank soil,

rocks within and outside the oasis56 (139.8) 51 (124.0)

10 Rangelands vegetated areas (grassland, bushes,

trees) in the proximity of the oasis 128 (43.4)

Classification based on NDVI thresholds only without the use of training sites

Fig. 4: Means of the spectral reflectance patterns of the five most important land use/land cover classes used for supervised

classification for the Landsat images of the Dzungarian River Basin, NW China, in 2008, 1999 and 1990.

the near-infrared (NIR) and red (RED) portions of the

electromagnetic spectrum (Tucker, 1979), and is calcu-

lated as NDVI = (NI R RE D)/(NIR +RE D). NDVI

values 0.05 were used to separate between vegetated

(class 10 = Rangeland) and non-vegetated rangeland ar-

eas (class 9 =Mountains).

To classify the panchromatic, historical Corona im-

ages a visual image interpretation was conducted

(Antrop & Van Eetvelde, 2000). An automatically ob-

ject oriented classification approach did not generatesatisfactory results, mainly because of the high hetero-

geneity in size, shape and colour of the ground ob-

jects within the classes (such as small irregular shaped

agricultural fields within the floodplains versus nearly

quadrate fields outside the floodplains) and similar

characteristics of ground objects belonging to different

classes (irregular spaced patches of grassland and arable

land in the floodplains). Some ground objects were too

small to be separated from neighbouring objects such as

single houses next to trees.

Since the detection of differences in vegetation coveris impossible in panchromatic data sets, vegetated and

8/9/2019 319-1040-2-PB.pdf

7/14


non-vegetated areas within the rangelands were not sep-

arated for the Corona images. Furthermore, classes

4 and 5 were combined to one class as it was not

detectable whether agricultural land was already har-

vested or not. Since salinity affected areas generally ap-

pear brighter than undisturbed agricultural land (Jensen,2007), we further reclassified very bright areas of the

agricultural land (raster value >225) to degraded land.

Visual inspections of the panchromatic Corona image

revealed, that some agricultural areas were located on

marginal sites characterised by shallow soils wich re-

stricted the water holding capacity. The interpretation

was based on the relief condition using a Digital Ele-

vation Model (NASAs Shuttle Radar Topography Mis-

sion; resolution = 90m) as ancillary information and the

spectral properties of the image, whereby dry soils gen-

erally cause a higher reflectance compared to wet soils

(Wilkie & Finn, 1996). Since we expected land aban-

donment to particularly occur in these areas, the land

use changes of the marginal agricultural areas were ex-

amined separately in an additional step of analysis (Ta-

ble 6).

2.4 Accuracy assessment

To identify the most suitable classification method for

the study area, an accuracy assessment was conducted

based on independent validation plots (Table 2) gathered

from high resolution images and own field observations.

Since no historical high resolution images were avail-

able, accuracy assessment using the kappa-coefficient

and an error matrix (Congalton & Green, 1999) was

only feasible for the year 2008. Based on the error ma-

trix the Producers Accuracy (PA), the Consumers Ac-

curacy (CA), the Error of Omission (EO) and the Er-

ror of Commission (EC) were calculated. The PA is a

measure to assess how correctly classes were classified,

whereas the CA reflects how reliable a classified map is.

The EO indicates which areas were wrongly excluded

from a class and the EC is a measure of which areas were

wrongly included in a class (Campbell, 2002). The ref-

erence data (Congalton & Green, 1999) for class 1 to 8

were identified using the pan-sharpened Landsat image

of 2008. For the classes 9 to 10 we used a HSV trans-

formed standard false colour composite of the 2008

Landsat dataset, suitable to discriminate vegetated andnon-vegetated areas (Sabins, 1997).

2.5 Post processing

For the most suitable classifier (Table 4), the clas-

sified Landsat data were post-processed to further im-

prove classification accuracy. Some classes had simi-

lar spectral characteristics, resulting in a high error of

confusion, particularly, non-vegetated mountainous re-

gions, urban areas, riverbanks and degraded agricultural

land. During post-processing the urbanized and agri-

culturally used oasis areas were reclassified. The pan-

sharpened Landsat, the Google Earth and the histori-

cal Corona satellite images allowed to visually identify

and digitize all urban areas from 1964-2008. Within this

potential urbanization zone, all areas misclassified as

Agricultural Land Degraded and Riverbanks were

reclassified as Urban Areas. Outside this zone, some

parts with sparsely vegetated arable land were misclas-

sified as Urban Areas and had thus to be reclassified as

Agricultural Land Crops.

To correct further misclassification errors, we de-

fined a zone of riverbanks along water streams using the

buffer function of ArcGIS 9.3 (vers. 1.1; ESRI, Red-

lands, CA, USA). The location of the streams was de-

fined through an automated extraction of the drainage

network from the digital elevation model using the

ArcHydro Tools contained in ArcGIS 9. According

to visual inspections, the maximum extension of river-

banks next to the streams was around 75 m. Within this

buffer zone all misclassified areas were reclassified into

Riverbanks. Outside this zone misclassified riverbanks

were corrected into the class Agricultural Land De-

graded (Table 3).

Table 3: Identified region of reclassification, type of reclassification and extend of reclassified area

(ha) for the post processing of the classified Landsat data (acquired in 1990, 1999 and 2008) in the

Dzungarian river basin, NW China. For class abbreviations see Table 2.

Extend of the Areas (ha)Region of Reclassification Type of Reclassification

1990 1999 2008

Inside urbanisation zone Class 3 to Class 1 15 5 38

Class 6 to Class 1 203 225 345

Outside urbanisation zone Class 1 to Class 5 489 221 528

Inside zone of riverbanks Class 1 to Class 3 127 38 222

Class 6 to Class 3 19 49 25

Class 9 to Class 3 90 97 67Outside zone of Riverbanks Class 3 to Class 6 67 60 84

8/9/2019 319-1040-2-PB.pdf

8/14


3 Results

3.1 Reliability of the determined sets of training sites

Reliability of the determined sets of training sites For

the Landsat images of the years 2008 and 1999 the

means of the spectral patterns were similar indicatingthat the training sites were identified correctly (Figure

4). They are slightly shifted due to different minimum

and maximum DN values. As the training sites for the

1990 Landsat dataset were identified in the same way as

the 1999 data, we assumed that they were also identified

correctly and that the differences of the spectral patterns

are only due to differences of the DN distribution.

3.2 Accuracy assessment of the classification results

The validation of the produced maps of the 2008 data

revealed that the best classification result was obtained

using the SMAPC method (Table 4). The subsequentpost processing to improve the accuracy of the classi-

fied urban areas and riverbanks increased the kappa co-

efficient from 0.90 to 0.92. The error matrices obtained

pointed to a major reduction of errors due to the post

processing (Table 5). Therefore this method was also

used for the classification of the 1990 and 1999 data.

Table 4:Accuracy assessment of the different classifiers (MLC

= Maximum Likelihood Classifier; SMAPC= Sequential A

Priori Classifier; Post Processing= post processing of the ur-

ban areas and riverbanks; Smooth = smoothing of the the-

matic map after classification and prior to the accuracy as-sessment) for the 2008 Landsat data used to examine LUCC

in the Dzungarian river basin, NW China.

Classifier Special Steps Kappa - Coefficient

SMAPC 0.8969

SMAPC Post Processing 0.9209

MLC Smoothing 0.8929

Some verified urban areas were misclassified as

Riverbanks, Agricultural Land Degraded and

Agricultural Land Crops. Post-processing, allowed

to reduce errors and confusion between urban areas and

riverbanks or degraded agricultural land. Therefore the

PA for urban areas strongly improved (Table 5b).

The high EO of the class Riverbanks was due to

misclassifications in Urban Areas and Mountains,

and due to a confusion with agricultural land. Post-

processing successfully diminished the former misclas-

sification error, thereby improving the PA from 57 % to

91%.

Even without post-processing, PAs and CAs of all

classes were > 90 %. Lower accuracies were only ob-

served for classes that depicted the more natural flood-

plain vegetation. The EO of these classes were mainly

due to misclassifications among each other and the ex-

clusion of areas misclassified as harvested or degraded

agricultural land as well as riverbanks. The PA of the

rangelands was < 90 % due to confusion with non-

vegetated areas near the oases. Nevertheless, a very high

CA of 100% was achieved for this class, which was sim-ilar to the classes Urban Areas and Water Bodies.

3.3 Changes of the different classes by surface area

The urban areas constantly expanded from 285 to

814 ha in 1964 and 2008 respectively (+186 %, Table

6). The total agricultural land increased by 61 % and

reached 6,834 ha in 2008. The highest increments oc-

curred between 1990 (5,000 ha) and 1999 (6,400 ha).

Compared to 1964, more land was classified as degraded

agricultural land in the following years, whereas in 1990

a maximum of 1,180 ha was detected. From 1990 to

2008 degraded farmland, which was apparently affectedby soil salinity decreased in size (630 ha). The area in

the floodplain covered with trees decreased until 1999

(-45 %) compared to 1964, but expanded by 12 % since

then. The grassland interspersed with bushes in the

floodplain slightly increased by 3 % between 1964 and

1990 but decreased constantly in the following years,

reaching a size of 1400 ha in 2008. The extent of the

water bodies fluctuated over the years with a minimum

of 280 ha and a maximum of 500 ha in 1964 and 1999

respectively. The total size of the oasis increased from

8,310 up to 9,160 ha during the studied time span. It

could also be determined that the rangelands were about

60 km2 smaller in 2008 than in 1990.

3.4 Land use changes of marginal agricultural areas

In 1964 around 19 % of the total agricultural land was

located on marginal sites characterized by shallow soils

and steep slopes. A small part of this land (5 %) was ur-

banised in the following years. The majority of this for-

mer agriculturally used land (30 %) was abandoned and

classified as non-vegetated mountainous region in the

years after 1964. Most of land abandonment occurred

in 1990, whereby the agricultural use at marginal sites

increased again in 1999 and 2008 by 760 ha up to 880ha. Conversely, the size of degraded agricultural land

increased from 240 ha in 1964 up to 490 ha in 1990

and declined until 2008, reaching an extent of 180 ha

(Table 7).

4 Discussion

4.1 LUCC in oases settlements and surrounding

rangelands

Overall, the trends of LUCC in mountain oases of the

present study were largely similar as compared to steppe

and deserted regions of northern China with decreasing

areas of grasslands and degraded land and expanding

8/9/2019 319-1040-2-PB.pdf

9/14

8/9/2019 319-1040-2-PB.pdf

10/14


Table 6: Post classification comparison showing the area (ha) of the different land uses classes in the Dzungarian river basin,

NW China (1964 2008). Calculated changes refer to 1964 for the classes of the oasis settlements and on 1990 for the classes of

the periphery.

1964 1990 1999 2008

Class

Area Area Changes (ha &%) Area Changes (ha &%) Area Changes (ha &%)

Urban Areas 285 646 361 (+127) 706 421 (+148) 814 529 (+186)

Water Bodies 280 344 65 (+23) 505 226 (+81) 366 86 (+31)

Riverbanks 69 258 189 (+272) 145 76 (+109) 333 264 (+381)

Agricultural Land-Harvested 3692 2583 1875 2668

Agricultural Land-Crops 1226 3802 3534

Agricultural Land-Degraded 544 1184 640 (+118) 720 176 (+32) 632 88 (+16)

Total Agriculture 4236 4993 757 (+18) 6397 2161 (+51) 6834 2598 (+61)

Floodplain Forest 1402 840 -562 (-40) 773 -629 (-45) 868 -534 (-38)

Floodplain Meadow 2676 2763 87 (+3) 1749 -927 (-35) 1455 -1221 (-46)Total Oases 8314 8596 282 (+3) 8919 605 (+7) 9158 843 (+10)

Mountains 27452 29210 1758 (+6) 32590 5138 (+19)

Rangeland 10860 8672 -2188 (-20) 4902 -5958 (-55)

Total Area 8314 48156 48157 48162

Table 7: Surface area (ha) and changes of cultivated and degraded agricultural land at marginal sites within oasis settlements in

the Dzungarian river basin, NWt China (1964 2008). Cultivated refers to agricultural land harvested as well as covered with

crops.

1964 1990 1999 2008

ClassArea Area Changes (ha &%) Area Changes (ha &%) Area Changes (ha &%)

Cultivated Agricultural Land 1346 545 -801 (-60) 760 -586 (-44) 880 -466 (-35)

Degraded Agricultural Land 236 492 255 (+108) 290 53 (+23) 183 -54 (-23)

During winter times livestock is kept in the floodplains

and the fodder mainly consists of hay and wheat straw

(Banks, 2001). Thus, it can be assumed that the impor-

tance of grasslands as potential fodder source declined

while crop residues (available at larger scales) becamemore important. The proportion of the area classified as

degraded agricultural land decreased from nearly 1,200

to 600 ha in 1990 and 2008 respectively; whereas the

level of degraded land was higher compared to 1964.

This indicates that the policy for reconverting degraded

land into pasture or forest land is already producing

measurable results. The expansion of woodland in the

floodplain by 12 % from 1999 to 2008 resulted from

reforestation efforts started in 2002. Management im-

provements have been undertaken as well to avoid soil

salinity and to reclaim degraded areas for agricultural

useas reported for other regions in northern China (Yuet al., 2010).

However, the differences in the acquisition dates of

the Landsat datasets may also be a reason for the de-

crease of degraded land. The Landsat image of 1990

was recorded one month later and thus around 20 %

more land was already harvested compared to 1999 and2008. As a result, more agricultural land was with-

out vegetation cover including land affected by low soil

salinity. These areas were not classified as degraded

in the images of 1999 and 2008, which were acquired

one month earlier, since the spectral information of the

cropped areas confounded the information of the soil

salinity. For a more precise estimation of the total ex-

tent of saline soils, imagery with a large amount of bare

soil taken outside the vegetation period during winter

times would reveal better results.

After 1964 a bigger area on the images was classi-

fied as Water Bodies. This accords well with the find-ings of Liet al.(2001) who reported that the water table

8/9/2019 319-1040-2-PB.pdf

11/14


within the oasis systems has been raised due to the large

scale application of irrigation water. The fluctuation of

the amount of irrigation water can be explained by dif-

ferences in the inflow of melt-water from the surround-

ing mountains, which feds most of the river systems in

northern China (Banks, 2001).According to the results in the present study, range-

lands decreased by nearly 60 km2 from 1990 to 2008.

Generally, the reduction of rangeland areas in arid and

semi-arid regions can be explained by two contrary

models. One is the non-equilibrium model, whereby the

extent and productivity of rangelands is mainly deter-

mined by precipitation (Sullivan & Rohde, 2002). The

equilibrium model in contrast identifies overgrazing as

a major cause for declining productivity and palatability

of the rangelands, through a reduction of biodiversity

and changes in species composition (Christensen et al.,

2003; Vetter, 2005). Overgrazing can make pasture land

more susceptible to wind erosion resulting in a decline

of rangeland areas (Baoping & Tianzong, 2001). For

our study area, the causes for the degradation of range-

land areas remain unclear since precipitation data does

not cover the whole time span and is only available un-

til 1997 (NCDC, 2009). Furthermore, it was impossi-

ble to diminish inter- and intra-annual effects of precip-

itation based on monthly NDVI calculations (Richard

& Poccard, 1998; Weiss et al., 2001), due to a lack of

sufficient Landsat datasets. Nonetheless, drought events

have been detected for 1989 and 1990. That similar re-

current drought events caused the observed reduction in

vegetation cover until 2008 is not very likely, indicatingthat overgrazing may also be responsible for the degra-

dation of the rangelands as explained by the equilibrium

model.

4.2 Evaluation of the conducted classification

Oasis systems in arid NW China are vulnerable micro

ecosystems encompassed by very large rangeland areas.

These two ecosystems differ substantially in produc-

tivity and species composition as it has been reported for

comparable regions in Mongolia (Hilbig & Tungalag,

2006). This is mainly due to differences in water avail-ability. The productivity of oasis systems is based on ir-

rigation water supplied by rivers fed by melt-water, the

surrounding rangelands instead depend on scarce pre-

cipitation events (Banks, 2001; Wang & Cheng, 2000).

Taking these characteristics into account, both ecosys-

tems were classified separately to gather more accurate

results for the analysis of LUCC. Furthermore, a pre-

study revealed various misclassifications between vege-

tated rangeland areas and land use classes of oases (such

as un-harvested agricultural land, grassland and bushes

of the floodplain). However, the separation of oasis set-

tlements and rangeland areas as an additional step dur-

ing the classification process is time consuming, partic-

ularly on a broader scale.

The SMAPC and the MLC method with subsequent

smoothing resulted in similar classification accuracies.

The former performed slightly better which agrees with

the findings of McCauley & Engel (1995) indicating that

this classifier is more efficient for classifying different

types of crops cultivated on homogenous fields. How-ever, a post processing of the classified Landsat data

was still necessary, as misclassification among urban ar-

eas, riverbanks, degraded agricultural land, and moun-

tainous areas occurred frequently. The post processing

of the classified Landsat datasets improved considerably

the PA values of the classes Urban Areas and River-

banks by 22 % and 34 %, respectively. Most of the land

use classes were classified in an acceptable way with

PA values > 90 %. On the other hand, post-processing

slightly increased the error of confusion among river-

banks, degraded agricultural land as well as agricul-

tural land covered with crops. Additionally, some wood-

land areas of the floodplains were misclassified as river-banks and agricultural land after post-processing. The

assumed zone of riverbanks used during post processing

still depends on some uncertainties, whereby the real ex-

tent of riverbanks was under- as well as overestimated

for some areas.

The use of ancillary information for a more powerful

and accurate automated classification is recommended.

For instance, Stefanov et al. (2001) used, among oth-

ers, the results of a texture analysis for the identification

of urban areas in a semi-arid region of Arizona, USA.

They reported that the occurrence of houses, streets and

paths caused more texture compared to homogenousagricultural fields. However, if a texture analysis suc-

cessfully identifies small-scale structures in the present

study (scattered houses in the floodplains, single bushes

or trees) remains questionable. To quantify soil salini-

sation more precisely, a correlation analysis between re-

flectance spectra of the Landsat datasets and measured

soil electric conductivity (EC) could be conducted. This

would facilitate a better identification of areas poten-

tially at risk of salinisation (Yuet al., 2010).

The low classification accuracies and misclassifica-

tions among trees, grassland interspersed with bushes

and agricultural used land can be explained by their sim-ilar reflectance characteristics, since all classes depict

highly productive landscape features. Another reason

is the patchy structure of the floodplains with small-

scale changes of different land use types. These may

enhance the classification of mixed pixels contain-

ing the spectral information of different ground features

(Wilkie & Finn, 1996). Hence, a fuzzy classification ap-

proach (Tso & Mather, 2001) would be more suitable

to classify these floodplain areas. Since each ground

object covered by one pixel contributes to the recorded

brightness value, the brightness of mixed pixels changes

slightly as the ratio of the covered ground objects vary.

By means of fuzzy rules and weighting factors, the re-searcher can define in which range of brightness values

8/9/2019 319-1040-2-PB.pdf

12/14


a mixed pixel should be dedicated as a member of a cer-

tain class (Tso & Mather, 2001). For instance, in the

present study mixed pixels contained information from

cultivated agricultural land and its surrounding vegeta-

tion (grass, bushes), whereat a corresponding range can

be applied to intensify the weighing of cultivated landand finally assess the biggest extension of farmland.

The specified NDVI threshold of 0.05 to classify the

studied rangelands revealed accurate classification re-

sults with a CA of 100 %. Wanget al. (2008) detected

even lower NDVI values for vegetated areas within arid

rangelands of NW China, but non-vegetated soils can

generate also slightly higher values than 0.05. How-

ever, reddish areas in a standard false colour image of

the 2008 Landsat dataset, indicating vegetation cover

(Sabins, 1997), were mostly concurrent to areas char-

acterised by NDVI values of 0.05 and higher.

For the panchromatic mode of the Corona images,

cumbersome visual image interpretation was necessary,

whereby harvested and non-harvested agricultural land

as well as vegetated and no-vegetated rangelands could

not be well differentiated and was thus not performed.

Even though visual interpretations are time consuming,

such methods facilitate the simultaneous processing of

texture, grey-colour level tones and geometric features,

and thus offer a rather accurate and realistic way of inter-

pretation (Ruelland et al., 2010). Furthermore, some ob-

served land cover changes may be overestimated as a re-

sult of possible biases in post-classification comparison

resulting from different classification approaches (visual

interpretation of Corona images versus automatic clas-sification of Landsat images).

5 Conclusions

Remote sensing methods for the analysis of land-

scape transformations processes are important monitor-

ing tools in fast developing countries, such as North-

ern China, where dramatic land use changes occurred in

the last 50 years. The current method used was adapted

to the specific landscape characteristics of the present

study area, whereas the SMAPC with subsequent post-processing revealed most accurate results. The extrap-

olation of this method is feasible for similar arid and

semi-arid environments, but for a broad scale applica-

tion the labour costs might be a limiting factor (digi-

tizing urban areas per hand, separation of oasis from

periphery), especially for the visual interpretation of

Corona images. However, for the detection of small-

scale changes within complex oasis systems this method

gives more reliable and accurate results compared to

standard classification methods.

The use of ancillary information during classification

process such as a digital elevation model, precipitation

data, agricultural statistics or the correlation between re-

flectance spectra of the Landsat datasets and measured

soil electric conductivity is recommended to increase

the classification accuracies of urban areas, abandoned

land and degraded land.

Acknowledgements

The authors are indebted to Prof. Ximing Zhang forhis tireless efforts in supporting the IFAD-funded WA-

TERCOPE project (I-R-1284) in the transborder Altay-

Dzungarian region of Mongolia and China.

References

Antrop, M. & Van Eetvelde, V. (2000). Holistic aspects

of suburban landscapes: visual images interpretation

and landscape metrices. Landscape and Urban Plan-

ning, 50, 4358.

Banks, T. (2001). Property rights and the environment inpastoral China: evidence from the field.Development

and Change, 32, 717740.

Baoping, S. & Tianzong, F. (2001). Desertification in

China and its control. In S. W. Breckle, M. Veste, &

W. Wucherer (Eds.),Sustainable Land Use in Deserts

(pp. 418426). Springer, Heidelberg.

Booth, W. H. & Oldfield, R. B. (1989). A compari-

son of classification algorithms in terms of speed and

accuracy after the application of a post-classification

modal filter. International Journal of Remote Sens-

ing, 10, 12711276.

Bouman, C. A. & Shapiro, M. (1994). A multiscale ran-dom field model for Bayesian image segmentation.

IEEE Transaction on Image Processing, 3 (2),162

177.

Bruelheide, H., Jandt, U., Gries, D., Thomas, F. M.,

Foetzki, A. & Buerkert, A. (2003). Vegetation

changes in a river oasis on the southern rim of the

Taklamakan Desert in China between 1956 and 2000.

Phytocoenologia, 33, 801818.

Brunner, P. A. (2005). Water and salt management in

the Yanqi Basin, China. Ph.D. thesis Eidgenossische

Technische Hochschule, Zurich.

Campbell, J. B. (2002). Introduction to remote sensing,

Third ed.. Guilford Press, New York.

di Castri, F. & Hadley, M. (1988). Enhancing the cred-

ibility of ecology: Interacting along and across hier-

achical scales. GeoJournal, 17, 535.

Chander, G., Markham, B. L. & Helder, D. L. (2009).

Summary of current radiometric calibration coeffi-

cients for Landsat MSS, TM, ETM+, and EO-1 ALI

sensors. Remote Sensing of Environment, 113, 893

903.

Christensen, L., Coughenour, M. B., Ellis, J. E. & Chen,

Z. Z. (2003). Sustainability of inner Mongolian grass-

lands: Application of the Savanna model. Journal of

Range Management, 56, 319327.

8/9/2019 319-1040-2-PB.pdf

13/14


Chuluun, T. & Ojima, D. (2002). Land use change and

carbon cycle in arid and semi-arid lands of East and

Central Asia. Science in China, (Series C) 45, 4854.

Congalton, R. G. & Green, K. (1999). Assessing the ac-

curacy of remotely sensed data. Principles and prac-

tices. CRC Press, Boca Raton.Gmaps Pedometer (2009). Calculation of the eleva-

tion. URL http://www.gmap-pedometer.com/

?r=2778274.

Graefe, S., Siebert, S., Bruelheide, H. & Buerkert, A.

(2004). Changes in the agricultural structure of the

Qira-Oasis. In M. Runge, & X. Zhang (Eds.),Con-

tributions to a Workshop on Sustainable Manage-

ment of the Shelterbelt Vegetation of River Oases in

the Taklimakan Desert. Urumqi 2003. Shaker Ver-

lag, Aachen.

Graetz, D. (1994). Grasslands. In W. Meyer, & B. L.

Turner (Eds.), Changes in land use and land cover,a global perspective(pp. 125148). Cambridge Uni-

versity Press, Cambridge.

Hao, H. & Ren, Z. (2009). Land use/land cover change

(LUCC) and eco-environment response to LUCC in

farming-pastoral zone, China. Agricultural Sciences

in China, 8, 9197.

Hilbig, W. & Tungalag, R. (2006). Vegetation-

skundliche Untersuchungen in der Borzongijn und

Galbyn-Gobi (Omnogov Aimak, Mongolei). Feddes

Repertorium, 117,399426.

Hill, J. (2001). Remote sensing of surface proper-

ties. The key to land degradation and desertifica-tion assessments. In S. W. Breckle, M. Veste, &

W. Wucherer (Eds.), Sustainable land use in deserts

(pp. 243254). Springer Verlag, Berlin.

Hu, Z.-Z., Yang, S. & Wu, R. (2003). Long-term cli-

mate variations in China and global warming signals.

Journal of Geophysical Research, 108 (D19). doi:

10.1029/2003JD003651.

Jensen, J. R. (2007).Remote sensing of the environment.

An earth resource perspective. Second. ed.. Pearson

Prentice Hall, New York.

Jia, B., Zhang, Z., Ci, L., Ren, Y., Pan, B. & Zhang, Z.

(2004). Oasis land-use dynamics and its influence onthe oasis environment in Xinjiang, China. Journal of

Arid Environments, 56, 1126.

Jun, L., Shixin, W. & Xiaolei, Z. (2006). Temporal-

spatial characteristics of the urbanization of rural land

in Xinjiang. Chinese Journal of Population, Re-

sources and Environment, 4, 3744.

Klein, J. A., Harte, J. & Zhao, X. Q. (2004). Experimen-

tal warming causes and rapid species loss, dampened

by simulated grazing, on the Tibetan Plateau.Ecology

Letters, 7, 11701179.

Li, X., Fang, J. & Zhough, H. F. (2001). The hydro-

logical effect under human activities in the inland wa-tersheds of Xinjiang, China. Chinese Geographical

Science, 11, 2734.

Li, X. L., Yuan, Q. H., Wan, L. Q. & He, F. (2008).

Perspective on livestock production systems in China.

Rangeland Journal, 30, 211220.

Li, Z., Li, X., Wang, Y., Ma, A. & Wang, J. (2004).

Land-use change analysis in Yulin prefecture, north-western China using remote sensing and GIS. Inter-

national Journal of Remote Sensing, 25, 56915703.

Lillesand, T. M. & Kiefer, R. W. (2000). Remote sens-

ing and image interpretation. Fourth ed.. Wiley, New

York.

Liu, Y., Zhang, X., Lei, J. & Zhu, L. (2010). Urban

expansion of oasis cities between 1990 and 2007 in

Xinjiang, China.International Journal of Sustainable

Development and World Ecology, 17, 253262.

Liu, Y. M. (1993). Rangeland degradation in the pastoral

region of China: causes and countermeasures.ACIAR

Techniqual Reports Series, 25, 2026.Luo, G. P., Zhou, C. H., Chen, X. & Li, Y. (2008).

A methodology of characterizing status and trend of

land changes in oases: A case study of Sangong River

watershed, Xinjiang, China. Journal of Environmen-

tal Management, 88, 775783.

McCauley, J. D. & Engel, B. A. (1995). Comparison of

scene segmentations: SMAP, ECHO and Maximum

Likelihood.IEEE Transactions on Geosience and Re-

mote Sensing, 33, 13131316.

Millward, J. A. & Tursun, N. (2004). Political history

and strategies of control, 1978 2001. In S. F. Starr

(Ed.),Xinjiang - Chinas Muslim Borderland(pp. 36101). M.E. Sharpe Inc., Armonk, New York.

NCDC (2009). Climate data of the Qinghe/Qinglii

station. URLhttp://www1.ncdc.noaa.gov/pub/

orders/CDO7563271849103.txt.

Oliver, C. & Quegan, S. (2004). Understanding syn-

thetic aperture radar images. Sci-Tech Publishing,

Raleigh.

Pal, M. & Mather, P. M. (2003). An assessment of

the effectiveness of decision tree methods for land

cover classification. Remote Sensing of Environment,

86, 554565.

Pohl, C. (1999). Tools and methods for fusion of images

of different spatial resolution. International Archives

of Photogrammetry and Remote Sensing, 32, H. Part

743 W6.

Qinghe County Administration (unpublished). Data ob-

tained from Qinghe County authorities in interviews

in 2007.

Quantum GIS Development Team (2009). Quantum

GIS (QGIS), Open Source Geospatial Foundation

(OSGeo), GNU General Public License. URL http:

//qgis.osgeo.org.

Richard, Y. & Poccard, I. (1998). A statistical study

of the NDVI sensitivity to seasonal and interannual

rainfall variations in Southern Africa. International

8/9/2019 319-1040-2-PB.pdf

14/14


Journal of Remote Sensing, 19, 29072920.

Ruelland, D., Levavasseur, F. & Tribotte, A. (2010). Pat-

terns and dynamics of land-cover changes since the

1960s over three experimental areas in Mali. Inter-

national Journal of Applied Earth Observation and

Geoinformation, 12, Supplement 1, S11S17. doi:10.1016/j.jag.2009.10.006.

Runnstrom, M. C. (2003). Rangeland development of

the Mu Us sandy land in semiarid China: an analysis

using Landsat and NOAA remote sensing data. Land

Degradation &Development, 14, 189202.

Sabins, F. F. (1997). Remote sensing. Principles and

interpretation. Third ed.. Freeman, New York.

Stefanov, L. W., Ramsey, M. S. & Christensen, P. R.

(2001). Monitoring urban land cover change: an ex-

pert system approach to land cover classification of

semiarid to arid urban centers.Remote Sensing of En-

vironment, 77, 173185.Sullivan, S. & Rohde, R. (2002). On non-equilibrium in

arid and semi-arid grazing systems. Journal of Bio-

geography, 29, 15951618.

Tso, B. & Mather, P. M. (2001). Classification methods

for remotely sensed data. Taylor & Francis, London.

Tucker, C. J. (1979). Red and photographic infrared lin-

ear combinations for monitoring vegetation. Remote

Sensing of Environment, 8, 127150.

Turner, M. G., Gardner, R. H. & ONeill, R. V. (2001).

Landscape ecology in theory and practise - Pattern

and Process. Springer, NY, Berlin, Hamburg.

Vetter, S. (2005). Rangelands at equilibrium and non-equilibrium: recent developments in the debate.Jour-

nal of Arid Environments, 62, 321341.

Wang, G. & Cheng, G. (2000). The characteristics of

water resources and the changes of the hydrological

process and environment in the arid zone of northwest

China. Environmental Geology, 39, 783790.

Wang, R., Fan, Z., Zhang, H., Chen, Y. & Ma, Y.

(2002). Remote sensing analysis of desert vegeta-

tion and its landscape changes: The case in middle

reaches of Tarim River Basin, Xinjiang, China. Sci-

ence in China, (Series D), 5458.

Wang, S., Liu, S., Wang, X., Guo, J., Shuai, Y. & Hu,

N. (2008). A simple method for information extrac-

tion of farm field in Tarim River Basin using MODIS

data. The International archives of the Photogram-

metry, Remote Sensing and Spatial Information Sci-

ences, 37, 10431046.

Weiss, E., Marsh, S. E. & Pfirman, E. S. (2001). Appli-

cation of NOAA-AVHRR NDVI time-series data to

assess changes in Saudi Arabias rangelands. Inter-national Journal of Remote Sensing, 22, 10051027.

Wiemer, C. (2004). The economy of Xinjiang. In S. F.

Starr (Ed.), Xinjiang - Chinas Muslim Borderland

(pp. 163189). M.E. Sharpe Inc., Armonk, New York.

Wilkie, D. S. & Finn, J. T. (1996). Remote sensingimagery for natural resources monitoring. A guide

for first-time users. Columbia University Press, New

York.

Xiaoa, J., Shen, Y., Ge, J., Tateishi, R., Tang, C., Liang,

Y. & Huang, Z. (2004). Evaluating urban expansion

and land use change in Shijiazhuang, China, by us-

ing GIS and remote sensing. Landscape and UrbanPlanning, 75, 6980.

Xiuling, F. S. C. (2001). Rationally utilizing water re-

sources to control soil salinity in irrigation districts.

In D. E. Stott, R. H. Mohtar, & G. C. Steinhardt

(Eds.), Sustaining the global farm, selected papers

from the 10th International Soil Conservation Orga-

nization Meeting (pp. 11341138). Purdue Univer-

sity and USDA-ARS National Soil Erosion Research

Laboratory.

Yu, R., Liu, T., Xu, Y., Zhu, C., Qu, Z., Liu, X. & Li, C.

(2010). Analysis of salinisation dynamics by remote

sensing in Hetao Irrigation District of North China.

Agricultural Water Management, 97, 19521960.

Zhang, K., Yu, Z., Li, X., Zhou, W. & Zhang, D. (2007).

Land use change and land degradation in China from

1991 to 2001. Land Degradation and Development,

18, 209219.

Zhang, X., Li, X. & Zhang, H. (2001). The control

of drift sand on the southern fringe of the Takla-

makan Desert and example from the Cele-oasis. In

S. Breckle, M. Veste, & W. Wucherer (Eds.),Sustain-

able land use in deserts(pp. 350356). Springer, Hei-

delberg.

Zhu, Z. D. & Chen, G. T. (1994).Sandy Desertificationin China. Status and Trends. China Science Press,

Beijing. 250pp. (in Chinese).

319-1040-2-PB.pdf

Documents

Transcript of 319-1040-2-PB.pdf